# Release-review browser handoff study: method and provenance

Study version: 2026-09-18.1. Recorded September 18, 2026. Publisher: Awesome Patel, SaaS Mirror, Chicago. Public article: https://saasmirror.com/research/release-review-handoff-study.

## Question

When an eighteen-check review is filtered to one visible follow-up, do its actual downloaded Markdown and JSON files still contain the complete record? What happens to a later unexported edit when navigating within the site?

## Browser and fixture

The baseline run used the live public https://saasmirror.com/tools/release-review in Chrome 153.0.0.0 on macOS 26.6.2, at 2560 by 1353 CSS pixels and device-pixel ratio 2. The CDP user-agent string advertised macOS 10.15.7; the actual host version was 26.6.2. The baseline was recorded at 2026-09-18T16:05:51Z. One primary export run was completed in one browser environment.

The release name was `FICTIONAL-fixture-01`. Check 01 had a fictional unresolved follow-up note; check 02 had a fictional pass marker that did not assess any product; check 03 had a fictional not-applicable reason. Fifteen entries remained unchecked. No real customer, client, employment, or private project data were entered.

## Steps and verification

1. Open a blank review and record its initial summary.
2. Enter the three fictional conditions, statuses, and notes.
3. Select **Needs follow-up** and record the visible row count and full summary.
4. Click **Export Markdown** and **Export JSON** in the browser. The interface only claims a download was requested. Independently inspect the files actually saved to disk.
5. Parse the downloaded JSON for eighteen unique check IDs, the expected status counts, and the three notes. Count eighteen check headings and the same three notes in the Markdown file. Preserve the original files and their hashes.
6. After exporting, edit the release field again, click the site's Home link, and revisit the tool. Record the URL, dialog events, and form state on return.
7. Separately, edit a fictional field and perform a full-page reload. Record whether the browser emits a `beforeunload` event without inferring what a human user would see.

The browser's Markdown `waitForEvent(download)` listener timed out. The 8,373-byte Markdown file appeared in Downloads at 16:06:23 UTC and its content/hash were verified. The 10,054-byte JSON file appeared at 16:06:50 UTC and was similarly verified. The offline validation passed: eighteen unique IDs, one follow-up, one pass, one not-applicable item, fifteen unchecked items, and the three fictional notes in both formats.

## Results and limitations

The selected filter showed one row; both saved files retained the complete eighteen-check record. After a later unexported edit, clicking an in-site Home link navigated away without a captured JavaScript dialog and returning showed a blank review. A full-page reload emitted one `beforeunload` event; browser automation handled it, so this does not establish that a person would have seen or accepted a prompt.

This is a test of this tool, not of another product's release. One browser run cannot establish behavior in other browsers, devices, assistive technologies, or network conditions. No network trace was captured, so browser-local processing was not independently established by this test. The live deployment's exact source commit was not proven by the browser record. Future corrections must retain this baseline and add separately dated observations rather than replacing it.

## Files and reuse

- `observations.json` and `observations.csv`: the seven recorded baseline steps. Their observation data are dedicated under CC0 1.0 (https://creativecommons.org/publicdomain/zero/1.0/).
- `filtered-followup.png` and `fresh-after-leave.png`: representative baseline screenshots.
- `saas-mirror-release-review-2026-09-18.json` and `.md`: the exact downloaded fictional review files. The JSON is 10,054 bytes with SHA-256 `62e2ef6032fb3e95aadf38944b2f0ed3628e27c506671e5bad7bdacb00b397d0`; the Markdown is 8,373 bytes with SHA-256 `5937b07af9ce37d00c9c42a1c6705e7fb7e1a12cd5629872c2451db0cfa235b9`.

The CC0 dedication covers only the observation JSON and CSV, not screenshots, article text, sample exports, or third-party references. Those materials follow the SaaS Mirror terms at https://saasmirror.com/terms.

## Navigation correction and local retest

The original live baseline is unchanged. A separately built local preview added a native HTML leave dialog for unexported edits. Five follow-up steps in `navigation-retest.json` observed that Home opened the dialog, **Keep editing** retained the same field value and URL, explicit discard navigated to Home, and a blank review left without a prompt. An export request cleared the warning state and allowed navigation; that follow-up step did not verify its downloaded file. The original export files above were independently verified.

`leave-dialog.png` is a 2048 by 1038 screenshot of the local retest with fictional input. This retest did not establish Back/Forward, tab closing, external-link, screen-reader, keyboard, mobile or cross-browser behavior. It verifies the tested internal-link path only. The tool does not save the review automatically, and a download request is not proof that a file was saved.

`navigation-retest.json` is included in the CC0 observation-data dedication. The correction is versioned beside the original baseline rather than replacing it.